[TD-52] [TD-148] add cluster test script

src/client/src/tscServer.c

@@ -163,11 +163,13 @@ void tscProcessActivityTimer(void *handle, void *tmrId) {
     tscGetQueryInfoDetailSafely(&pSql->cmd, 0, &pQueryInfo);
     pQueryInfo->command = TSDB_SQL_HB;
     
+    pSql->cmd.command = TSDB_SQL_HB;
     if (TSDB_CODE_SUCCESS != tscAllocPayload(&(pSql->cmd), TSDB_DEFAULT_PAYLOAD_SIZE)) {
       tfree(pSql);
       return;
     }
 
+    pSql->cmd.command = TSDB_SQL_HB;
     pSql->param = pObj;
     pSql->pTscObj = pObj;
     pSql->signature = pSql;

src/client/src/tscSystem.c

@@ -67,7 +67,6 @@ int32_t tscInitRpc(const char *user, const char *secret) {
     rpcInit.label = "TSC-vnode";
     rpcInit.numOfThreads = tscNumOfThreads;
     rpcInit.cfp = tscProcessMsgFromServer;
-    rpcInit.ufp = tscUpdateIpSet;
     rpcInit.sessions = tsMaxVnodeConnections;
     rpcInit.connType = TAOS_CONN_CLIENT;
     rpcInit.user = (char*)user;
@@ -96,6 +95,7 @@ int32_t tscInitRpc(const char *user, const char *secret) {
     rpcInit.label = "TSC-mgmt";
     rpcInit.numOfThreads = 1;
     rpcInit.cfp = tscProcessMsgFromServer;
+    rpcInit.ufp = tscUpdateIpSet;
     rpcInit.sessions = tsMaxMgmtConnections;
     rpcInit.connType = TAOS_CONN_CLIENT;
     rpcInit.idleTime = 2000;

src/dnode/src/dnodeMClient.c

@@ -59,6 +59,19 @@ void dnodeUpdateIpSet(void *ahandle, SRpcIpSet *pIpSet) {
   tsMnodeIpSet = *pIpSet;
 }
 
+void dnodeGetMnodeIpSet(void *ipSetRaw, bool usePublicIp) {
+  SRpcIpSet *ipSet = ipSetRaw;
+  ipSet->numOfIps = tsMnodeInfos.nodeNum;
+  ipSet->inUse = tsMnodeInfos.inUse;
+  for (int32_t i = 0; i < tsMnodeInfos.nodeNum; ++i) {
+    if (usePublicIp) {
+      ipSet->ip[i] = tsMnodeInfos.nodeInfos[i].nodeIp;
+    } else {
+      ipSet->ip[i] = tsMnodeInfos.nodeInfos[i].nodeIp;
+    }
+  }
+}
+
 int32_t dnodeInitMClient() {
   dnodeReadDnodeCfg();
   tsRebootTime = taosGetTimestampSec();
@@ -138,7 +151,9 @@ static void dnodeProcessRspFromMnode(SRpcMsg *pMsg) {
   if (tsDnodeProcessMgmtRspFp[pMsg->msgType]) {
     (*tsDnodeProcessMgmtRspFp[pMsg->msgType])(pMsg);
   } else {
-    dError("%s is not processed in mnode rpc client", taosMsg[pMsg->msgType]);
+    dError("%s is not processed in dnode mclient", taosMsg[pMsg->msgType]);
+    SRpcMsg rpcRsp = {.pCont = 0, .contLen = 0, .code = TSDB_CODE_OPS_NOT_SUPPORT, .handle = pMsg->handle};
+    rpcSendResponse(&rpcRsp);
   }
 
   rpcFreeCont(pMsg->pCont);

src/dnode/src/dnodeMnode.c

@@ -88,7 +88,7 @@ static void dnodeProcessMsgFromMnode(SRpcMsg *pMsg) {
   if (dnodeProcessMgmtMsgFp[pMsg->msgType]) {
     (*dnodeProcessMgmtMsgFp[pMsg->msgType])(pMsg);
   } else {
-    dError("%s is not processed in mserver", taosMsg[pMsg->msgType]);
+    dError("%s is not processed in dnode mserver", taosMsg[pMsg->msgType]);
     rspMsg.code = TSDB_CODE_MSG_NOT_PROCESSED;
     rpcSendResponse(&rspMsg);
     rpcFreeCont(pMsg->pCont);

src/inc/dnode.h

@@ -41,8 +41,9 @@ void *dnodeAllocateRqueue(void *pVnode);
 void  dnodeFreeRqueue(void *rqueue);
 void  dnodeSendRpcWriteRsp(void *pVnode, void *param, int32_t code);
 
-bool  dnodeIsFirstDeploy();
+bool     dnodeIsFirstDeploy();
 uint32_t dnodeGetMnodeMasteIp();
+void     dnodeGetMnodeIpSet(void *ipSet, bool usePublicIp);
 void *   dnodeGetMnodeInfos();
 int32_t  dnodeGetDnodeId();
 

src/mnode/src/mgmtDClient.c

@@ -79,7 +79,9 @@ static void mgmtProcessRspFromDnode(SRpcMsg *rpcMsg) {
   if (mgmtProcessDnodeRspFp[rpcMsg->msgType]) {
     (*mgmtProcessDnodeRspFp[rpcMsg->msgType])(rpcMsg);
   } else {
-    mError("%s is not processed in dclient", taosMsg[rpcMsg->msgType]);
+    mError("%s is not processed in mgmt dclient", taosMsg[rpcMsg->msgType]);
+    SRpcMsg rpcRsp = {.pCont = 0, .contLen = 0, .code = TSDB_CODE_OPS_NOT_SUPPORT, .handle = rpcMsg->handle};
+    rpcSendResponse(&rpcRsp);
   }
 
   rpcFreeCont(rpcMsg->pCont);

src/mnode/src/mgmtDServer.c

@@ -108,8 +108,12 @@ static void mgmtProcessMsgFromDnode(SRpcMsg *rpcMsg) {
     bool usePublicIp = false;
     
     SRpcIpSet ipSet = {0};
-    mgmtGetMnodeIpSet(&ipSet, usePublicIp);
-    mTrace("conn from dnode ip:%s redirect msg", taosIpStr(connInfo.clientIp));
+    ipSet.port = tsMnodeDnodePort;
+    dnodeGetMnodeIpSet(&ipSet, usePublicIp);
+    mTrace("conn from dnode ip:%s user:%s redirect msg, inUse:%d", taosIpStr(connInfo.clientIp), connInfo.user, ipSet.inUse);
+    for (int32_t i = 0; i < ipSet.numOfIps; ++i) {
+      mTrace("index:%d ip:%s", i, taosIpStr(ipSet.ip[i]));
+    }
     rpcSendRedirectRsp(rpcMsg->handle, &ipSet);
     return;
   }
@@ -119,7 +123,8 @@ static void mgmtProcessMsgFromDnode(SRpcMsg *rpcMsg) {
     memcpy(pMsg, rpcMsg, sizeof(SRpcMsg));
     mgmtAddToDServerQueue(pMsg);
   } else {
-    mError("%s is not processed in dserver", taosMsg[rpcMsg->msgType]);
+    mError("%s is not processed in mgmt dserver", taosMsg[rpcMsg->msgType]);
+    mgmtSendSimpleResp(rpcMsg->handle, TSDB_CODE_MSG_NOT_PROCESSED);
     rpcFreeCont(rpcMsg->pCont);
   }
 }

src/mnode/src/mgmtShell.c

@@ -151,8 +151,13 @@ static void mgmtProcessMsgFromShell(SRpcMsg *rpcMsg) {
     bool usePublicIp = (connInfo.serverIp == tsPublicIpInt);
     
     SRpcIpSet ipSet = {0};
-    mgmtGetMnodeIpSet(&ipSet, usePublicIp);
-    mTrace("conn from ip:%s user:%s redirect msg", taosIpStr(connInfo.clientIp), connInfo.user);
+    ipSet.port = tsMnodeShellPort;
+    dnodeGetMnodeIpSet(&ipSet, usePublicIp);
+    mTrace("conn from shell ip:%s user:%s redirect msg, inUse:%d", taosIpStr(connInfo.clientIp), connInfo.user, ipSet.inUse);
+    for (int32_t i = 0; i < ipSet.numOfIps; ++i) {
+      mTrace("index:%d ip:%s", i, taosIpStr(ipSet.ip[i]));
+    }
+
     rpcSendRedirectRsp(rpcMsg->handle, &ipSet);
     return;
   }
@@ -429,9 +434,9 @@ static void mgmtProcessConnectMsg(SQueuedMsg *pMsg) {
 connect_over:
   rpcRsp.code = code;
   if (code != TSDB_CODE_SUCCESS) {
-    mLError("user:%s login from %s, code:%d", connInfo.user, taosIpStr(connInfo.clientIp), code);
+    mLError("user:%s login from %s, result:%s", connInfo.user, taosIpStr(connInfo.clientIp), tstrerror(code));
   } else {
-    mLPrint("user:%s login from %s, code:%d", connInfo.user, taosIpStr(connInfo.clientIp), code);
+    mLPrint("user:%s login from %s, result:%s", connInfo.user, taosIpStr(connInfo.clientIp), tstrerror(code));
     rpcRsp.pCont   = pConnectRsp;
     rpcRsp.contLen = sizeof(SCMConnectRsp);
   }
@@ -488,7 +493,7 @@ static bool mgmtCheckMsgReadOnly(SQueuedMsg *pMsg) {
 }
 
 static void mgmtProcessUnSupportMsg(SRpcMsg *rpcMsg) {
-  mError("%s is not processed in shell", taosMsg[rpcMsg->msgType]);
+  mError("%s is not processed in mnode shell", taosMsg[rpcMsg->msgType]);
   SRpcMsg rpcRsp = {
     .msgType = 0,
     .pCont   = 0,

tests/script/sh/deploy.sh

@@ -93,7 +93,7 @@ echo "privateIp           $NODE_IP"       >> $TAOS_CFG
 echo "dDebugFlag          199"            >> $TAOS_CFG
 echo "mDebugFlag          199"            >> $TAOS_CFG
 echo "sdbDebugFlag        199"            >> $TAOS_CFG
-echo "rpcDebugFlag        131"            >> $TAOS_CFG
+echo "rpcDebugFlag        135"            >> $TAOS_CFG
 echo "tmrDebugFlag        131"            >> $TAOS_CFG
 echo "cDebugFlag          135"            >> $TAOS_CFG
 echo "httpDebugFlag       131"            >> $TAOS_CFG

tests/script/unique/mnode/mgmt33.sim

@@ -116,7 +116,7 @@ sleep 8000
 
 sql show mnodes
 $dnode1Role = $data3_1
-$dnode2Role = $data3_2
+$dnode2Role = $data3_4
 $dnode3Role = $data3_3
 print 192.168.0.1 ==> $dnode1Role
 print 192.168.0.2 ==> $dnode2Role
@@ -138,13 +138,13 @@ sleep 10000
 
 sql show mnodes
 $dnode1Role = $data3_1
-$dnode2Role = $data3_2
+$dnode2Role = $data3_4
 $dnode3Role = $data3_3
 print 192.168.0.1 ==> $dnode1Role
 print 192.168.0.2 ==> $dnode2Role
 print 192.168.0.3 ==> $dnode3Role
 
-if $dnode1Role != undecided then
+if $dnode1Role != offline then
   return -1
 endi
 #if $dnode2Role != master then

tests/script/unique/mnode/mgmt34.sim

@@ -183,7 +183,7 @@ endi
 
 print ============== step7
 system sh/exec_up.sh -n dnode1 -s stop
-sleep 10000
+sleep 4000
 
 sql show mnodes
 $dnode1Role = $data3_1
@@ -195,7 +195,7 @@ print 192.168.0.2 ==> $dnode2Role
 print 192.168.0.3 ==> $dnode3Role
 print 192.168.0.4 ==> $dnode4Role
 
-if $dnode1Role != undecided then
+if $dnode1Role != offline then
   return -1
 endi
 
@@ -205,7 +205,7 @@ sleep 8000
 
 sql show mnodes
 $dnode1Role = $data3_1
-$dnode2Role = $data3_2
+$dnode2Role = $data3_5
 $dnode3Role = $data3_3
 $dnode4Role = $data3_4
 print 192.168.0.1 ==> $dnode1Role